drm/nouveau/ttm: untangle code to support accelerated buffer moves

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

8 files changed, 119 insertions, 94 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 6d66314d16bd..6e78b1aaa74d 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -36,6 +36,7 @@
 #include "nouveau_mm.h"
 #include "nouveau_vm.h"
 #include "nouveau_fence.h"
+#include "nouveau_ramht.h"
 
 #include <linux/log2.h>
 #include <linux/slab.h>
@@ -511,6 +512,17 @@ nve0_bo_move_copy(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 }
 
 static int
+nvc0_bo_move_init(struct nouveau_channel *chan, u32 handle)
+{
+	int ret = RING_SPACE(chan, 2);
+	if (ret == 0) {
+		BEGIN_NVC0(chan, NvSubCopy, 0x0000, 1);
+		OUT_RING  (chan, handle);
+	}
+	return ret;
+}
+
+static int
 nvc0_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 		  struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
 {
@@ -528,17 +540,17 @@ nvc0_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 		if (ret)
 			return ret;
 
-		BEGIN_NVC0(chan, NvSubM2MF, 0x0238, 2);
+		BEGIN_NVC0(chan, NvSubCopy, 0x0238, 2);
 		OUT_RING  (chan, upper_32_bits(dst_offset));
 		OUT_RING  (chan, lower_32_bits(dst_offset));
-		BEGIN_NVC0(chan, NvSubM2MF, 0x030c, 6);
+		BEGIN_NVC0(chan, NvSubCopy, 0x030c, 6);
 		OUT_RING  (chan, upper_32_bits(src_offset));
 		OUT_RING  (chan, lower_32_bits(src_offset));
 		OUT_RING  (chan, PAGE_SIZE); /* src_pitch */
 		OUT_RING  (chan, PAGE_SIZE); /* dst_pitch */
 		OUT_RING  (chan, PAGE_SIZE); /* line_length */
 		OUT_RING  (chan, line_count);
-		BEGIN_NVC0(chan, NvSubM2MF, 0x0300, 1);
+		BEGIN_NVC0(chan, NvSubCopy, 0x0300, 1);
 		OUT_RING  (chan, 0x00100110);
 
 		page_count -= line_count;
@@ -550,6 +562,28 @@ nvc0_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 }
 
 static int
+nv50_bo_move_init(struct nouveau_channel *chan, u32 handle)
+{
+	int ret = nouveau_notifier_alloc(chan, NvNotify0, 32, 0xfe0, 0x1000,
+					 &chan->m2mf_ntfy);
+	if (ret == 0) {
+		ret = RING_SPACE(chan, 6);
+		if (ret == 0) {
+			BEGIN_NV04(chan, NvSubCopy, 0x0000, 1);
+			OUT_RING  (chan, handle);
+			BEGIN_NV04(chan, NvSubCopy, 0x0180, 3);
+			OUT_RING  (chan, NvNotify0);
+			OUT_RING  (chan, NvDmaFB);
+			OUT_RING  (chan, NvDmaFB);
+		} else {
+			nouveau_ramht_remove(chan, NvNotify0);
+		}
+	}
+
+	return ret;
+}
+
+static int
 nv50_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 		  struct ttm_mem_reg *old_mem, struct ttm_mem_reg *new_mem)
 {
@@ -573,7 +607,7 @@ nv50_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 			if (ret)
 				return ret;
 
-			BEGIN_NV04(chan, NvSubM2MF, 0x0200, 7);
+			BEGIN_NV04(chan, NvSubCopy, 0x0200, 7);
 			OUT_RING  (chan, 0);
 			OUT_RING  (chan, 0);
 			OUT_RING  (chan, stride);
@@ -586,7 +620,7 @@ nv50_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 			if (ret)
 				return ret;
 
-			BEGIN_NV04(chan, NvSubM2MF, 0x0200, 1);
+			BEGIN_NV04(chan, NvSubCopy, 0x0200, 1);
 			OUT_RING  (chan, 1);
 		}
 		if (old_mem->mem_type == TTM_PL_VRAM &&
@@ -595,7 +629,7 @@ nv50_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 			if (ret)
 				return ret;
 
-			BEGIN_NV04(chan, NvSubM2MF, 0x021c, 7);
+			BEGIN_NV04(chan, NvSubCopy, 0x021c, 7);
 			OUT_RING  (chan, 0);
 			OUT_RING  (chan, 0);
 			OUT_RING  (chan, stride);
@@ -608,7 +642,7 @@ nv50_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 			if (ret)
 				return ret;
 
-			BEGIN_NV04(chan, NvSubM2MF, 0x021c, 1);
+			BEGIN_NV04(chan, NvSubCopy, 0x021c, 1);
 			OUT_RING  (chan, 1);
 		}
 
@@ -616,10 +650,10 @@ nv50_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 		if (ret)
 			return ret;
 
-		BEGIN_NV04(chan, NvSubM2MF, 0x0238, 2);
+		BEGIN_NV04(chan, NvSubCopy, 0x0238, 2);
 		OUT_RING  (chan, upper_32_bits(src_offset));
 		OUT_RING  (chan, upper_32_bits(dst_offset));
-		BEGIN_NV04(chan, NvSubM2MF, 0x030c, 8);
+		BEGIN_NV04(chan, NvSubCopy, 0x030c, 8);
 		OUT_RING  (chan, lower_32_bits(src_offset));
 		OUT_RING  (chan, lower_32_bits(dst_offset));
 		OUT_RING  (chan, stride);
@@ -628,7 +662,7 @@ nv50_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 		OUT_RING  (chan, height);
 		OUT_RING  (chan, 0x00000101);
 		OUT_RING  (chan, 0x00000000);
-		BEGIN_NV04(chan, NvSubM2MF, NV_MEMORY_TO_MEMORY_FORMAT_NOP, 1);
+		BEGIN_NV04(chan, NvSubCopy, NV_MEMORY_TO_MEMORY_FORMAT_NOP, 1);
 		OUT_RING  (chan, 0);
 
 		length -= amount;
@@ -639,6 +673,24 @@ nv50_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 	return 0;
 }
 
+static int
+nv04_bo_move_init(struct nouveau_channel *chan, u32 handle)
+{
+	int ret = nouveau_notifier_alloc(chan, NvNotify0, 32, 0xfe0, 0x1000,
+					 &chan->m2mf_ntfy);
+	if (ret == 0) {
+		ret = RING_SPACE(chan, 4);
+		if (ret == 0) {
+			BEGIN_NV04(chan, NvSubCopy, 0x0000, 1);
+			OUT_RING  (chan, handle);
+			BEGIN_NV04(chan, NvSubCopy, 0x0180, 1);
+			OUT_RING  (chan, NvNotify0);
+		}
+	}
+
+	return ret;
+}
+
 static inline uint32_t
 nouveau_bo_mem_ctxdma(struct ttm_buffer_object *bo,
 		      struct nouveau_channel *chan, struct ttm_mem_reg *mem)
@@ -661,7 +713,7 @@ nv04_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 	if (ret)
 		return ret;
 
-	BEGIN_NV04(chan, NvSubM2MF, NV_MEMORY_TO_MEMORY_FORMAT_DMA_SOURCE, 2);
+	BEGIN_NV04(chan, NvSubCopy, NV_MEMORY_TO_MEMORY_FORMAT_DMA_SOURCE, 2);
 	OUT_RING  (chan, nouveau_bo_mem_ctxdma(bo, chan, old_mem));
 	OUT_RING  (chan, nouveau_bo_mem_ctxdma(bo, chan, new_mem));
 
@@ -673,7 +725,7 @@ nv04_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 		if (ret)
 			return ret;
 
-		BEGIN_NV04(chan, NvSubM2MF,
+		BEGIN_NV04(chan, NvSubCopy,
 				 NV_MEMORY_TO_MEMORY_FORMAT_OFFSET_IN, 8);
 		OUT_RING  (chan, src_offset);
 		OUT_RING  (chan, dst_offset);
@@ -683,7 +735,7 @@ nv04_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 		OUT_RING  (chan, line_count);
 		OUT_RING  (chan, 0x00000101);
 		OUT_RING  (chan, 0x00000000);
-		BEGIN_NV04(chan, NvSubM2MF, NV_MEMORY_TO_MEMORY_FORMAT_NOP, 1);
+		BEGIN_NV04(chan, NvSubCopy, NV_MEMORY_TO_MEMORY_FORMAT_NOP, 1);
 		OUT_RING  (chan, 0);
 
 		page_count -= line_count;
@@ -743,16 +795,7 @@ nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr,
 			goto out;
 	}
 
-	if (dev_priv->card_type < NV_50)
-		ret = nv04_bo_move_m2mf(chan, bo, &bo->mem, new_mem);
-	else
-	if (dev_priv->card_type < NV_C0)
-		ret = nv50_bo_move_m2mf(chan, bo, &bo->mem, new_mem);
-	else
-	if (dev_priv->card_type < NV_E0)
-		ret = nvc0_bo_move_m2mf(chan, bo, &bo->mem, new_mem);
-	else
-		ret = nve0_bo_move_copy(chan, bo, &bo->mem, new_mem);
+	ret = dev_priv->ttm.move(chan, bo, &bo->mem, new_mem);
 	if (ret == 0) {
 		ret = nouveau_bo_move_accel_cleanup(chan, nvbo, evict,
 						    no_wait_reserve,
@@ -764,6 +807,42 @@ out:
 	return ret;
 }
 
+void
+nouveau_bo_move_init(struct nouveau_channel *chan)
+{
+	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
+	static const struct {
+		const char *name;
+		u32 oclass;
+		int (*exec)(struct nouveau_channel *,
+			    struct ttm_buffer_object *,
+			    struct ttm_mem_reg *, struct ttm_mem_reg *);
+		int (*init)(struct nouveau_channel *, u32 handle);
+	} _methods[] = {
+		{  "COPY", 0xa0b5, nve0_bo_move_copy, nvc0_bo_move_init },
+		{  "M2MF", 0x9039, nvc0_bo_move_m2mf, nvc0_bo_move_init },
+		{  "M2MF", 0x5039, nv50_bo_move_m2mf, nv50_bo_move_init },
+		{  "M2MF", 0x0039, nv04_bo_move_m2mf, nv04_bo_move_init },
+		{}
+	}, *mthd = _methods;
+	const char *name = "CPU";
+	int ret;
+
+	do {
+		ret = nouveau_gpuobj_gr_new(chan, mthd->oclass, mthd->oclass);
+		if (ret == 0) {
+			ret = mthd->init(chan, mthd->oclass);
+			if (ret == 0) {
+				dev_priv->ttm.move = mthd->exec;
+				name = mthd->name;
+				break;
+			}
+		}
+	} while ((++mthd)->exec);
+
+	NV_INFO(chan->dev, "MM: using %s for buffer copies\n", name);
+}
+
 static int
 nouveau_bo_move_flipd(struct ttm_buffer_object *bo, bool evict, bool intr,
 		      bool no_wait_reserve, bool no_wait_gpu,
@@ -920,8 +999,8 @@ nouveau_bo_move(struct ttm_buffer_object *bo, bool evict, bool intr,
 		goto out;
 	}
 
-	/* Software copy if the card isn't up and running yet. */
-	if (!dev_priv->channel) {
+	/* CPU copy if we have no accelerated method available */
+	if (!dev_priv->ttm.move) {
 		ret = ttm_bo_move_memcpy(bo, evict, no_wait_reserve, no_wait_gpu, new_mem);
 		goto out;
 	}
diff --git a/drivers/gpu/drm/nouveau/nouveau_channel.c b/drivers/gpu/drm/nouveau/nouveau_channel.c
index 9420538d2374..629d8a2df5bd 100644
--- a/drivers/gpu/drm/nouveau/nouveau_channel.c
+++ b/drivers/gpu/drm/nouveau/nouveau_channel.c
@@ -355,7 +355,7 @@ nouveau_channel_ref(struct nouveau_channel *chan,
 	*pchan = chan;
 }
 
-void
+int
 nouveau_channel_idle(struct nouveau_channel *chan)
 {
 	struct drm_device *dev = chan->dev;
@@ -370,6 +370,7 @@ nouveau_channel_idle(struct nouveau_channel *chan)
 
 	if (ret)
 		NV_ERROR(dev, "Failed to idle channel %d.\n", chan->id);
+	return ret;
 }
 
 /* cleans up all the fifos from file_priv */
diff --git a/drivers/gpu/drm/nouveau/nouveau_dma.h b/drivers/gpu/drm/nouveau/nouveau_dma.h
index b8838592d4e9..8db68be9544f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_dma.h
+++ b/drivers/gpu/drm/nouveau/nouveau_dma.h
@@ -48,13 +48,12 @@ void nv50_dma_push(struct nouveau_channel *, struct nouveau_bo *,
 
 /* Hardcoded object assignments to subchannels (subchannel id). */
 enum {
-	NvSubM2MF	= 0,
+	NvSubCtxSurf2D  = 0,
 	NvSubSw		= 1,
-	NvSubCtxSurf2D  = 2,
+	NvSubImageBlit  = 2,
 	NvSub2D		= 3,
 	NvSubGdiRect    = 3,
 	NvSubCopy	= 4,
-	NvSubImageBlit  = 4
 };
 
 /* Object handles. */
@@ -74,6 +73,7 @@ enum {
 	NvSema		= 0x8000000f,
 	NvEvoSema0	= 0x80000010,
 	NvEvoSema1	= 0x80000011,
+	NvNotify1       = 0x80000012,
 
 	/* G80+ display objects */
 	NvEvoVRAM	= 0x01000000,
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index 1ede35491f54..634d222c93de 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -693,6 +693,9 @@ struct drm_nouveau_private {
 		struct ttm_bo_global_ref bo_global_ref;
 		struct ttm_bo_device bdev;
 		atomic_t validate_sequence;
+		int (*move)(struct nouveau_channel *,
+			    struct ttm_buffer_object *,
+			    struct ttm_mem_reg *, struct ttm_mem_reg *);
 	} ttm;
 
 	struct {
@@ -930,7 +933,7 @@ extern void nouveau_channel_put_unlocked(struct nouveau_channel **);
 extern void nouveau_channel_put(struct nouveau_channel **);
 extern void nouveau_channel_ref(struct nouveau_channel *chan,
 				struct nouveau_channel **pchan);
-extern void nouveau_channel_idle(struct nouveau_channel *chan);
+extern int  nouveau_channel_idle(struct nouveau_channel *chan);
 
 /* nouveau_object.c */
 #define NVOBJ_ENGINE_ADD(d, e, p) do {                                         \
@@ -1322,6 +1325,7 @@ extern int nv04_crtc_create(struct drm_device *, int index);
 
 /* nouveau_bo.c */
 extern struct ttm_bo_driver nouveau_bo_driver;
+extern void nouveau_bo_move_init(struct nouveau_channel *);
 extern int nouveau_bo_new(struct drm_device *, int size, int align,
 			  uint32_t flags, uint32_t tile_mode,
 			  uint32_t tile_flags,
diff --git a/drivers/gpu/drm/nouveau/nouveau_object.c b/drivers/gpu/drm/nouveau/nouveau_object.c
index d7e56ce410b0..a3f961d98992 100644
--- a/drivers/gpu/drm/nouveau/nouveau_object.c
+++ b/drivers/gpu/drm/nouveau/nouveau_object.c
@@ -508,7 +508,6 @@ nouveau_gpuobj_gr_new(struct nouveau_channel *chan, u32 handle, int class)
 		return eng->object_new(chan, oc->engine, handle, class);
 	}
 
-	NV_ERROR(dev, "illegal object class: 0x%x\n", class);
 	return -EINVAL;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_state.c b/drivers/gpu/drm/nouveau/nouveau_state.c
index e4e73a13a2b2..9b4c900a3f30 100644
--- a/drivers/gpu/drm/nouveau/nouveau_state.c
+++ b/drivers/gpu/drm/nouveau/nouveau_state.c
@@ -509,73 +509,16 @@ nouveau_card_channel_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_channel *chan;
-	int ret, oclass;
+	int ret;
 
 	ret = nouveau_channel_alloc(dev, &chan, NULL, NvDmaFB, NvDmaTT);
 	dev_priv->channel = chan;
 	if (ret)
 		return ret;
-
 	mutex_unlock(&dev_priv->channel->mutex);
 
-	if (dev_priv->card_type <= NV_50) {
-		if (dev_priv->card_type < NV_50)
-			oclass = 0x0039;
-		else
-			oclass = 0x5039;
-
-		ret = nouveau_gpuobj_gr_new(chan, NvM2MF, oclass);
-		if (ret)
-			goto error;
-
-		ret = nouveau_notifier_alloc(chan, NvNotify0, 32, 0xfe0, 0x1000,
-					     &chan->m2mf_ntfy);
-		if (ret)
-			goto error;
-
-		ret = RING_SPACE(chan, 6);
-		if (ret)
-			goto error;
-
-		BEGIN_NV04(chan, NvSubM2MF, NV_MEMORY_TO_MEMORY_FORMAT_NAME, 1);
-		OUT_RING  (chan, NvM2MF);
-		BEGIN_NV04(chan, NvSubM2MF, NV_MEMORY_TO_MEMORY_FORMAT_DMA_NOTIFY, 3);
-		OUT_RING  (chan, NvNotify0);
-		OUT_RING  (chan, chan->vram_handle);
-		OUT_RING  (chan, chan->gart_handle);
-	} else
-	if (dev_priv->card_type <= NV_D0) {
-		ret = nouveau_gpuobj_gr_new(chan, 0x9039, 0x9039);
-		if (ret)
-			goto error;
-
-		ret = RING_SPACE(chan, 2);
-		if (ret)
-			goto error;
-
-		BEGIN_NVC0(chan, NvSubM2MF, 0x0000, 1);
-		OUT_RING  (chan, 0x00009039);
-	} else
-	if (dev_priv->card_type <= NV_E0) {
-		/* not used, but created to get a graph context */
-		ret = nouveau_gpuobj_gr_new(chan, 0xa040, 0xa040);
-		if (ret)
-			goto error;
-
-		/* bind strange copy engine to subchannel 4 (fixed...) */
-		ret = RING_SPACE(chan, 2);
-		if (ret)
-			goto error;
-
-		BEGIN_NVC0(chan, NvSubCopy, 0x0000, 1);
-		OUT_RING  (chan, 0x0000a0b5);
-	}
-
-	FIRE_RING (chan);
-error:
-	if (ret)
-		nouveau_card_channel_fini(dev);
-	return ret;
+	nouveau_bo_move_init(chan);
+	return 0;
 }
 
 static const struct vga_switcheroo_client_ops nouveau_switcheroo_ops = {
diff --git a/drivers/gpu/drm/nouveau/nv50_fbcon.c b/drivers/gpu/drm/nouveau/nv50_fbcon.c
index 61747e0d1180..e3c8b05dcae4 100644
--- a/drivers/gpu/drm/nouveau/nv50_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nv50_fbcon.c
@@ -201,8 +201,7 @@ nv50_fbcon_accel_init(struct fb_info *info)
 
 	BEGIN_NV04(chan, NvSub2D, 0x0000, 1);
 	OUT_RING(chan, Nv2D);
-	BEGIN_NV04(chan, NvSub2D, 0x0180, 4);
-	OUT_RING(chan, NvNotify0);
+	BEGIN_NV04(chan, NvSub2D, 0x0184, 3);
 	OUT_RING(chan, chan->vram_handle);
 	OUT_RING(chan, chan->vram_handle);
 	OUT_RING(chan, chan->vram_handle);
diff --git a/drivers/gpu/drm/nouveau/nve0_graph.c b/drivers/gpu/drm/nouveau/nve0_graph.c
index 41356274ca61..8a8051b68f10 100644
--- a/drivers/gpu/drm/nouveau/nve0_graph.c
+++ b/drivers/gpu/drm/nouveau/nve0_graph.c
@@ -822,7 +822,7 @@ nve0_graph_create(struct drm_device *dev)
 	NVOBJ_CLASS(dev, 0xa0c0, GR); /* subc 1: COMPUTE */
 	NVOBJ_CLASS(dev, 0xa040, GR); /* subc 2: P2MF */
 	NVOBJ_CLASS(dev, 0x902d, GR); /* subc 3: 2D */
-	//NVOBJ_CLASS(dev, 0xa0b5, GR); /* subc 4: COPY */
+	NVOBJ_CLASS(dev, 0xa0b5, GR); /* subc 4: COPY */
 	return 0;
 
 error: